Image recognition: face recognition

This article is reproduced from the WeChat public account "Living in the Information Age". The author lives in the information age. To reprint this article, please contact the Living in the Information Age public account.

For image recognition, the most popular application field is face recognition. It is the large-scale application of facial recognition technology that gives our country's Sky Eye Project and Xueliang Project, which are spread in every corner, a greater application space, and also makes our country safer. Next, let’s take a look at the development history of face recognition applications.

September 2017. Apple held its autumn conference and launched the iPhone Face recognition is essentially a type of image recognition. It is an identification technology based on people's facial feature information. Use a camera or video camera to collect images or video streams containing human faces, and automatically detect and track faces in the images, and then perform a series of face-related technologies on the detected faces, which are usually also called portrait recognition and facial recognition.

Face recognition began in the 1960s, with the development of computer technology and optical imaging technology. It truly entered the primary application stage in the late 1990s, with technology from the United States, Japan and Germany. Lord. With the development of artificial intelligence and rapid iterative updates of processing, facial recognition technology has also achieved great breakthroughs. At the same time, facial recognition is also the latest application of biometrics. The realization of its core technology demonstrates the transformation from weak artificial intelligence to strong artificial intelligence. In general, the principle of face recognition is to collect the user's facial data and store it in a database, and then perform machine learning to collect the facial data of the object that needs to be unlocked, put it into the database for comparison, and finally complete the unlocking.

The appearance of face recognition on mobile phones is not Apple’s first. Google added the face recognition function in Android 4.0, but at that time The technology does not have security features. At that time, face recognition was mainly limited by two factors: 1. The mobile phone did not have enough space to stack more advanced face recognition sensors; 2. The algorithm had a bottleneck, which was also the key point, and could not calculate the uneven texture of the face. , only stays in the 2d plane stage. The iPhoneX’s small notch integrates 8 sensors, 4 of which serve face recognition. In fact, Apple had already begun to lay out 3D depth sensing components earlier and acquired an Israeli facial recognition company in advance.

The new generation iPhone Combined with an infrared lens to read the depth reflected by these light points, the 3D structure of the face can be quickly scanned, and combined with 3D modeling technology to complete the collection and recognition of facial feature information. Because the human face is not flat, the geometric accuracy of the 3D human facial data collected by Face ID will be very high, greatly reducing the error rate. Moreover, the speed of Face ID facial recognition function is much faster than that of Touch ID fingerprint recognition.

Abroad, in addition to Apple, Sony and Samsung have demonstrated 3D facial recognition technology. In China, Huawei mobile phones subsequently launched the Honor V9, a facial recognition 3D modeling mobile phone, and Xiaomi's note3 is also equipped with face unlocking black technology, etc. These all show that 3D facial recognition technology has become a trend in future mobile phone development. In addition, Alibaba also applied facial recognition to Alipay, and successfully unlocked facial recognition payment after fingerprint payment. In March 2019, at the IT exhibition held in Hannover, Germany, Jack Ma presented a speech to German Chancellor Merkel and China Vice Prime Minister Ma Kai demonstrated Alipay's "face recognition" payment, which aroused heated discussions about facial recognition technology.

Alipay’s face recognition is also based on deep learning, that is, the collected images are preprocessed first, that is, key feature point detection, rotation, normalization of the distance between the eyes, and image cutting, etc. The method performs face alignment and then uses different scales for multi-channel normalization. Multiple face information is intercepted at key feature points for learning. Deep learning uses multi-layer convolutional neural network learning to extract the features of this area in each face area. CNN has three cores: local perception, full value sharing, and time or Spatial subsampling, the combination of these three ideas ensures that shift, scale and deformation invariance are obtained to a certain extent. Finally, a classifier is used to determine whether they are the same person.

Tencent Cloud Shentu·Face Recognition (Face Recognition) is based on Tencent Youtu's powerful facial analysis technology, providing services including face detection and analysis, facial features positioning, face search, face comparison, face verification, personnel With multiple functions such as duplication checking and liveness detection, it provides developers and enterprises with high-performance and high-availability face recognition services. It can be used in a variety of application scenarios such as smart retail, smart communities, online entertainment, smart buildings, and online identity authentication to fully meet the needs of customers in various industries for facial attribute recognition and user identity confirmation.

Its advantages are as follows:

(1) Accurate recognition: Tencent Cloud face recognition service has multiple It set new records in international public competitions. The accuracy of face comparison in the 2017 LFW evaluation was as high as 99.80%. Face search ranked first in the million-scale MegaFace competition with a recognition rate of 83.29%, leading the industry in recognition accuracy.

(2) Algorithm leadership: Based on the third-generation Tencent Youtu grandmother model, it integrates metric learning, transfer learning, multi-task learning and other training methods to optimize the model; customized according to the characteristics of different business scenarios Fine-tuning or distilling model meets the dual requirements of business performance and latency.

(3) Stable and reliable: Tencent Cloud face recognition service has been verified by massive users and complex scenarios of Tencent’s internal products. It operates stably and robustly, with service availability exceeding 99.9%.

(4) Real-time response: Face recognition has the characteristics of high concurrency, high throughput, and low latency. Even a million-scale face search can still be processed in just hundreds of milliseconds, satisfying your needs. real-time usage requirements.

(5) Simple and easy to use: Provides a rich and diverse online API and offline identification SDK, which not only supports access to the cloud, but also supports the offline identification SDK to be embedded into applications and hardware devices. You can also use it as needed The offline identification SDK and online API are combined to form a device-cloud solution to meet the needs of different scenarios.

(6) Wide application: Face recognition is widely used in many scenarios such as online photo albums, smart retail, sensitive face review, face access control, face attendance, face login, face special effects, online exams, etc. .

In addition, Baidu, Google, etc. have also done a lot of research and products on face recognition. Facial recognition technology is becoming increasingly mature, bringing great convenience to our work and life.

The above is the detailed content of Image recognition: face recognition. For more information, please follow other related articles on the PHP Chinese website!